Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Librispeech] Add 'all' config #4184

Merged
merged 10 commits into from
Apr 22, 2022
Merged

[Librispeech] Add 'all' config #4184

Changes from 2 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
9bfbc14
[Librispeech] Add 'all' config
Apr 19, 2022
fe68645
Update datasets/librispeech_asr/librispeech_asr.py
patrickvonplaten Apr 19, 2022
189ee52
apply suggestions
Apr 20, 2022
2f11a71
Fix merge
Apr 20, 2022
627c99f
correct paths
Apr 20, 2022
31e67cb
up
Apr 20, 2022
3944a3b
up
Apr 20, 2022
5f699fa
up
Apr 21, 2022
2adb2d7
up
Apr 21, 2022
bce8221
up
Apr 21, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
136 changes: 96 additions & 40 deletions datasets/librispeech_asr/librispeech_asr.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,6 +57,12 @@ def map_to_array(batch):
_URL = "http://www.openslr.org/12"
_DL_URL = "http://www.openslr.org/resources/12/"


def retrieve_subset(name):
# "train-other-500.tar.gz" -> "other-500"
return "-".join(name.split(".")[0].split("-")[1:])


_DL_URLS = {
"clean": {
"dev": _DL_URL + "dev-clean.tar.gz",
Expand All @@ -69,6 +75,14 @@ def map_to_array(batch):
"dev": _DL_URL + "dev-other.tar.gz",
"train.500": _DL_URL + "train-other-500.tar.gz",
},
"all": {
"dev": {retrieve_subset(u): _DL_URL + u for u in ["dev-clean.tar.gz", "dev-other.tar.gz"]},
patrickvonplaten marked this conversation as resolved.
Show resolved Hide resolved
"test": {retrieve_subset(u): _DL_URL + u for u in ["test-clean.tar.gz", "test-other.tar.gz"]},
patrickvonplaten marked this conversation as resolved.
Show resolved Hide resolved
"train": {
patrickvonplaten marked this conversation as resolved.
Show resolved Hide resolved
retrieve_subset(u): _DL_URL + u
for u in ["train-clean-100.tar.gz", "train-clean-360.tar.gz", "train-other-500.tar.gz"]
},
},
}


Expand All @@ -94,6 +108,7 @@ class LibrispeechASR(datasets.GeneratorBasedBuilder):
BUILDER_CONFIGS = [
LibrispeechASRConfig(name="clean", description="'Clean' speech."),
LibrispeechASRConfig(name="other", description="'Other', more challenging, speech."),
LibrispeechASRConfig(name="all", description="Combined clean and other dataset."),
]

def _info(self):
Expand All @@ -107,6 +122,7 @@ def _info(self):
"speaker_id": datasets.Value("int64"),
"chapter_id": datasets.Value("int64"),
"id": datasets.Value("string"),
"subset": datasets.Value("string"),
}
),
supervised_keys=("file", "text"),
Expand All @@ -117,61 +133,101 @@ def _info(self):

def _split_generators(self, dl_manager):
archive_path = dl_manager.download(_DL_URLS[self.config.name])

if self.config.name == "clean":
train_splits = [
datasets.SplitGenerator(
name="train.100", gen_kwargs={"files": dl_manager.iter_archive(archive_path["train.100"])}
name="train.100",
gen_kwargs={"files": {"clean-100": dl_manager.iter_archive(archive_path["train.100"])}},
),
datasets.SplitGenerator(
name="train.360", gen_kwargs={"files": dl_manager.iter_archive(archive_path["train.360"])}
name="train.360",
gen_kwargs={"files": {"clean-360": dl_manager.iter_archive(archive_path["train.360"])}},
),
]
dev_splits = [
datasets.SplitGenerator(
name=datasets.Split.VALIDATION,
gen_kwargs={"files": {"clean": dl_manager.iter_archive(archive_path["dev"])}},
)
]
test_splits = [
datasets.SplitGenerator(
name=datasets.Split.TEST,
gen_kwargs={"files": {"clean": dl_manager.iter_archive(archive_path["test"])}},
)
]
elif self.config.name == "other":
train_splits = [
datasets.SplitGenerator(
name="train.500", gen_kwargs={"files": dl_manager.iter_archive(archive_path["train.500"])}
),
name="train.500",
gen_kwargs={"files": {"other-500": dl_manager.iter_archive(archive_path["train.500"])}},
)
]
dev_splits = [
datasets.SplitGenerator(
name=datasets.Split.VALIDATION,
gen_kwargs={"files": {"other": dl_manager.iter_archive(archive_path["dev"])}},
)
]
test_splits = [
datasets.SplitGenerator(
name=datasets.Split.TEST,
gen_kwargs={"files": {"other": dl_manager.iter_archive(archive_path["test"])}},
)
]
elif self.config.name == "all":
train_splits = [
datasets.SplitGenerator(
name="train",
gen_kwargs={"files": {k: dl_manager.iter_archive(v) for k, v in archive_path["train"].items()}},
)
]
dev_splits = [
datasets.SplitGenerator(
name=datasets.Split.VALIDATION,
gen_kwargs={"files": {k: dl_manager.iter_archive(v) for k, v in archive_path["dev"].items()}},
)
]
test_splits = [
datasets.SplitGenerator(
name=datasets.Split.TEST,
gen_kwargs={"files": {k: dl_manager.iter_archive(v) for k, v in archive_path["test"].items()}},
)
]

return train_splits + [
datasets.SplitGenerator(
name=datasets.Split.VALIDATION, gen_kwargs={"files": dl_manager.iter_archive(archive_path["dev"])}
),
datasets.SplitGenerator(
name=datasets.Split.TEST, gen_kwargs={"files": dl_manager.iter_archive(archive_path["test"])}
),
]
return train_splits + dev_splits + test_splits

def _generate_examples(self, files):
"""Generate examples from a LibriSpeech archive_path."""
key = 0
audio_data = {}
transcripts = []
for path, f in files:
if path.endswith(".flac"):
id_ = path.split("/")[-1][: -len(".flac")]
audio_data[id_] = f.read()
elif path.endswith(".trans.txt"):
for line in f:
if line:
line = line.decode("utf-8").strip()
id_, transcript = line.split(" ", 1)
audio_file = f"{id_}.flac"
speaker_id, chapter_id = [int(el) for el in id_.split("-")[:2]]
transcripts.append(
{
"id": id_,
"speaker_id": speaker_id,
"chapter_id": chapter_id,
"file": audio_file,
"text": transcript,
}
)
if audio_data and len(audio_data) == len(transcripts):
for transcript in transcripts:
audio = {"path": transcript["file"], "bytes": audio_data[transcript["id"]]}
yield key, {"audio": audio, **transcript}
key += 1
audio_data = {}
transcripts = []
for subset, iterator in files.items():
for path, f in iterator:
if path.endswith(".flac"):
id_ = path.split("/")[-1][: -len(".flac")]
audio_data[id_] = f.read()
elif path.endswith(".trans.txt"):
for line in f:
if line:
line = line.decode("utf-8").strip()
id_, transcript = line.split(" ", 1)
audio_file = f"{id_}.flac"
speaker_id, chapter_id = [int(el) for el in id_.split("-")[:2]]
transcripts.append(
{
"id": id_,
"speaker_id": speaker_id,
"chapter_id": chapter_id,
"file": audio_file,
"text": transcript,
"subset": subset,
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@albertz @lhoestq - this "subset" arg is one of "clean-100", "clean-360", "other-500" for:

ds = load_dataset("librispeech_asr", "all", split="train")
ds[0]

This should allow the user to easily define which subsets are needed I think

}
)
if audio_data and len(audio_data) == len(transcripts):
for transcript in transcripts:
audio = {"path": transcript["file"], "bytes": audio_data[transcript["id"]]}
yield key, {"audio": audio, **transcript}
key += 1
audio_data = {}
transcripts = []